基于改进BiLSTM的ADS-B信号欺骗检测方法研究

Study on the deception detection method of ADS-B signals based on improved BiLSTM

摘要:

随着智能交通系统的快速发展,自动相关监视-广播(Automatic Dependent Surveillance-Broadcast,ADS-B)技术作为一种先进的空中交通管理监控手段得到了广泛应用。然而,ADS-B信号的开放性和易受攻击性使其成为潜在的欺骗攻击目标。为了提高飞行安全性,避免ADS-B系统受到欺骗式干扰,研究提出了一种基于深度学习的全向信标信号处理方法,用于检测ADS-B信号中的欺骗行为。研究利用全向信标收集ADS-B信号数据并提取相关特征,随后应用双向长短期记忆网络(Bidirectional Long Short-Term Memory,BiLSTM)深度学习模型对提取的特征进行训练,以区分正常信号和欺骗信号。然后结合焦点损失函数和贝叶斯优化算法对信号检测方法进行优化,并通过几何位置相关函数量化飞行状态误差。结果表明,模型的训练损失值和训练准确率分别达到了0.25和98.15%,改进后的BiLSTM模型在分类性能上所有指标均超过了99.50%。此外,研究方法在飞行速度、水平飞行方向和垂直飞行方向的检测误差分别仅为0.01%、0.01%和0.04%。对真实信号的检测显示,其飞行速度、水平和垂直方向的损失值均为1,而欺骗信号在这些指标上的损失值误差分别为15%、1%和0.3%。综上所述,面向全向信标信号处理的深度学习ADS-B信号欺骗检测方法研究,有效实现了优异的检测准确率和鲁棒性,为民用航空安全领域提供了重要的技术支持与参考。

Abstract:

With the rapid development of intelligent transportation systems, Automatic Dependent Surveillance Broadcast (ADS-B) technology has been widely used as an advanced means of air traffic management and monitoring. However, the openness and vulnerability of ADS-B signals make them potential targets for deception attacks. In order to improve flight safety and avoid deceptive interference in ADS-B systems, a deep learning based omnidirectional beacon signal processing method is proposed to detect deceptive behavior in ADS-B signals. Research on using omnidirectional beacons to collect ADS-B signal data and extract relevant features, and then applying a Bidirectional Long Short Term Memory (BiLSTM) deep learning model to train the extracted features to distinguish between normal signals and deception signals. Then, the signal detection method is optimized by combining the focus loss function and Bayesian optimization algorithm, and the flight state error is quantified through the geometric position correlation function. The results showed that the training loss value and training accuracy of the model reached 0.25 and 98.15%, respectively. The improved BiLSTM model achieved classification performance of over 99.50% in all indicators. In addition, the detection errors of the research method in flight speed, horizontal flight direction, and vertical flight direction are only 0.01%, 0.01%, and 0.04%, respectively. The detection of real signals shows that the loss values of flight speed, horizontal and vertical directions are all 1, while the loss errors of deception signals on these indicators are 15%, 1% and 0.3%, respectively. In summary, the research on deep learning ADS-B signal deception detection methods for omnidirectional beacon signal processing has effectively achieved excellent detection accuracy and robustness, providing important technical support and reference for the field of civil aviation safety.